INTERNATIONAL JOURNALOF WIRELESS AND MOBILE COMPUTING
1
Modeling Synchronization of a Coordination System Anthony Y. Chang, Jason C. Hung and Chuan-Ho Kao
Abstract— This paper develops a general, theoretical computational model for discussing synchronization and specification scheme. We analyze the domains of relationships between agents in a coordination system. A set of algorithms is proposed to derive reasonable relations between agents. Possible conflicts in the agent specification are firstly detected and eliminated. The mechanism constructs partial order relations among actions. We also propose a temporal algebra to deal with qualitative and quantitative temporal relationships and to reason about definite and indefinite time. The algebra can also integrate instant points and even intervals. Several computation tables are proposed, and each table include a set of complete logics. The mechanism is efficiently to eliminate conflict specification and to generate synchronization scenarios. The contributions of this paper are using the generic relationship framework to handle synchronization specifications, and to eliminate conflicts in order to satisfy agent request. Index Terms— Agent Synchronization, Multi -Agent, Scheduling, Coordination Systems, Interval Algebra
I. INTRODUCTION
W
hen several agents work together, it is necessary to communicate between the master agent and the other worker agent. To synchronize the various types of agents is the major challenge for a coordination system. The coordination of actions is the set of supplementary activities which need to be carried out in a multi-agent environment[1]. When accessing common resources, in order to guarantee the system remains in a coherent state, actions have to be synchronized by computing procedures[2]. Biniaris present key issues related to the distributed implementation of a Finite-Difference Time Domain (FDTD) code using java mobile agents, and special agent communication and synchronization aspects related to FDTD Anthony Y. Chang is with the Department of Information Technology, the Overseas Chinese Institute of Technology , Taichung, Taiwan, R.O.C. (corresponding author, phone, fax: +886-4-2701-6855 x1508; e-mail: achang@ ocit.edu.tw). Jason C. Hung is with the Department of Information Management, Northern Taiwan Institute of Science and Technology , Taipei, Taiwan, R.O.C.(e-mail:
[email protected]). Chuan-Ho Kao is a PhD candidate of Computer Science and Information Engineering at Tamkang University, Taipei Hsien, Taiwan, R.O.C.(e-mail:
[email protected] ).
are presented in [3]. Conservative and optimistic approaches for resolving distributed simulation of multi-agent hybrid systems is presented in [4] Karim Hussein[7] provided a collaborative agent interaction and synchronization system for insuring effective structuring and control of a distributed conference. Shivakant Mishra and Peng Xie[5] design an interagent communication and synchronization model in the DaAgent mobile agent-based computing system. There are some temporal frameworks to specify temporal data model of synchronization. Little and Ghafoor[6] proposed the object composition Petri Net(OCPN). However it does not easily capture the distributed nature of application or react to the real world, just indicated in [13] The time-flow graph (TFG) is proposed in [14] but the graphs fail to represent interactive semantics and to reduce ambiguous situations. Time-line diagramis a n general and intuitive model, such as QuickTime[15] and MAEStro[16]. In a time-line, it could compute the absolute and relative time precisely, but it fails to specify uncertain and indefinite temporal information. A. Safavi and S. F. Smith[10] proposed a search-based approaches to measure the global impact of a commitment on the entire schedule. We analysis the relationship between agents and define a temporal algebra system for unifying and scheduling based on our early research result about synchronization mechanisms [2]. A set of algorithms is proposed to derive consistent relations between agents. The mechanism is efficiently to eliminate conflict specification and to generate synchronization constraints.
II. CONFLICTS OF A COORDINATION SYSTEM A coordination system usually contains several agents must be coordinated and a number of resources must be shared and synchronized with actions. These resources need to be arranged as layout. Modeling an agent temporal scenario often requires synchronizing the distributed resources . In following sections, we propose an integrated temporal computational model to deal with the following inconsistencies. l Qualitative inconsistency: the conflicts occur in the semantics or logics of temporal relations. Such as a scenario plays dependently with other objects which can not be satisfied with temporal relations. Moreover events with unpredictable times of occurrence and durations may
INTERNATIONAL JOURNALOF WIRELESS AND MOBILE COMPUTING
l
l
l
violate some specification constraints. For example, media A, B plays before media C in a container followed a specification , there exist inconsistency between temporal specifications. Quantitative inconsistency: the conflicts occur in the scheduling on syncbase-value, event-value, or offset-value. Resources limitation: resources are needed to accomplish agent actions, they are limited and vehicles are obliged to coordinate their action to avoid each other. Resources have to be shared with eliminating pointless actions, reducing costs and avoiding any possible conflicts. Mutual Constraints: the conflicts occur in the interdependencies between actions of agents.
III. MULTI -AGENT S YNCHRONIZATION J. Feber[2] indicated that rapidity, adaptability and predictiveness are the temporal characteristics of the coordination system. Also, to synchronize several actions have to define the manner in which actions are time-related. Coordination is a matter of positioning actions in time and space. As soon as several agents have to move together, their movements have to be synchronized in time and location. Since the agent synchronization base on time with a dynamic variation, the inconsistency often occurs in both qualitative semantics and quantitative values. We analyze conflicts between actions of a set of agents, and propose a methodology based on temporal algebra to deal with qualitative and quantitative inconsistency. It not only eliminates conflicts relations but also reasons about indefinite, uncertain, and incomplete temporal relationships.
IV. RELATIONAL DOMAIN This Section describes the symbolic constraint propagation. In a coordination system, it is necessary to manage the inter-agent communications between the actions of a set of agents. Three agents have to take mutual dependencies and constraints. The general idea is to use the existing information about the relations among time intervals or instants to derive the composition relations. For example, there existsthree agents X, Y, and Z, have to coordinate their action, it means that not only must X coordinate with Y, but also coordinate with Z. With specifying interdependencies, if "X before Y" and "Y before Z", it plains action of agent X has to be before Z. The composition may result in a multiple derivation. For example, if “X before Y” and “Y during Z”, the composed relation for X and Z could be “before”, “overlaps”, “meets”, “during”, or “starts”. If the composed relation could be any one of some relations, these derived relations are calledreasonable relations in our discussion. A reasonable set is a set of reasonable relations according to our definition. A reasonable set can not be
2
empty, since there must exist at least one relation between any two events, assuming that they are in the same one dimension (i.e. the time line). In some cases, relation compositions may result in a conflict specification due to the user specification or involved events synchronously. For example, if specifications "X before Y", "Y before Z", and "X after Z" are declared by the user, there exists a conflict between X and Z. When the specific relations are not found in derived reasonable set, the specification may cause conflicts. We analyze the domain of temporal relations and use a directed graph to compute the relations of agents. In the computation, we consider all possibilities: the unknown derivations, the multiple derivations, and the conflict derivations. Definition : An user edge denotes a relation between a pair of objects defined by the user. The relation may be reasonable or non-reasonable. Definition : A derived edge holds a non-empty set of reasonable relations derived by our algorithm. The relation of the two objects connected by the derived edge can be any reasonable relation in the set. For each pair of objects in the time line, there exists a set of possible binary relations held between the pair of objects. For an arbitrary number of objects (denoted by nodes), some of the relations (denoted by edges) are specified by the user while others are derived. If there exists a cycle in the directed relation graph, a conflict derivation may occur. If there exists no cycle, there is no conflict. Based on the above considerations, we suggest that the computation domain reveals four types, as discussed below. The complete relation domain is a complete graph which contains possible conflicts. We want to find a reasonable relation domain containing no conflict derivation. Note that, in these two domains (i.e., the complete and the reasonable), both user edges and derived edges exist. If there is a conflict among a set of user edges, one of the user edge must be removed from the cycle, or the relation of that user edge must be re-assigned. If there is no conflict, the two domains are equal. The reduced relation domain contains relations specified by the specification only. It is possible that the user issues a conflict situation. To avoid the occurrence of conflicts, we place a restriction on the user’s interaction. Instead of allowing the user to add an arbitrary relation to the relation graph, we only allow the user to add objects to a restricted relation domain, which is a tree and a sub-domain of the reduced relation domain. That is, when the user is about to add a new edge, the user either adds a new node connected to an existing node via an user edge, or joins two sub-trees via the user edge. No cycle is created in the restricted relation domain. Thus, the conflict situation does not exist. When deleting an user edge, the user has to maintain
INTERNATIONAL JOURNALOF WIRELESS AND MOBILE COMPUTING the connectivity of the tree. If all nodes are connected, the user specification is complete. Otherwise, the coordination system should alert the user to complete the specification. The above domains can be summarized as the following: l The complete relation domain (a complete graph) contains user edges and derived edges, with possible cycles and possible conflicts. l The reasonable relation domain (a graph) contains user edges and derived edges, with possible cycles but no conflict. l The reduced relation domain (a graph) contains only user
restricte
reasonabl
reduce
complete
Fig. 1. Relational Domain
qualitative interval relations. For solving point/interval algebra networks, we develop an O(n 2)-time algorithm that is an O(n) improvement for finding all pairs of feasible relations. Reasoning qualitative temporal knowledge is less precise than reasoning quantitative knowledge. Since the data are imprecise, the conclusions will be imprecise also. In many situations, imprecise conclusions are sufficient or they are the best users can hope for. However, it is not sufficient to label precise values for some computing systems. Applications that reason with partial temporal information require the ability to represent quantitative and qualitative relations among time points, time intervals and durations. This chapter integrates qualitative and quantitative information to deal with both definite and uncertain temporal specification. Figure 3 gives an example to represent an interval relation with endpoint constraints qualitatively. Given two intervals A and B, if end of A is before end of B, i.e. Ae²Be = {
